factors determining maritime transport market: time series
TRANSCRIPT
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 375
Factors Determining Maritime Transport Market:
Time Series Econometric Analysis on Container
Throughput of Cambodia’s Ports
KiriromCheav, PhD
Graduation School, National University of Management
Phnom Penh, Cambodia
Abstract:-The significantly increased demand for
container throughput and cargo traffic volume
transported in Cambodia has improved and accelerated
the port output advancement and the development of new
terminal. The main objective of this research is to
investigate the long-run and short-run relationships, as
well as to identify the influences on Cambodian combined
international ports’ container throughput.The researcher
collected the yearly data (1992 - 2017) from reliable and
official sources of PAS and PPAP ports, IMF, WB, and
MOC, and additionally interviewed with port authorities,
port operators, and qualified stakeholders. The
econometric approaches of The Unit Roots Test, Johansen
Test of Cointegration, VECM, and Test for Stability are
performed and applied in the analysis on the time series
data between container throughput as the dependent
variable and independent variables such as number of
ship call, export, import, and GDP per capita. The
estimation result confirmed and explained the effect of
long-run relationships between container throughput and
number of ship call, export, import, and GDP per capita
for one rank of vector cointegration. That means all the
factors move together to explain the changing demand for
container throughput of Cambodia’s ports in the long-run
equilibrium. The number of ship call and export has a
significantly positive impact. Alternatively, the import and
GDP per capita shows a statistically significant and
relatedly negative influence on container throughput.
Nevertheless, it is crucial to disclose the estimation
resultfrom the VECM did not find a short-run
relationship. That means, in the short-run equilibrium, the
affecting factors did not move together to demonstrate the
change of container throughput demand. It is crucial for
policy implication for government, port authorities, etc.
The best attentive insights, measures and actions would
help manage and achieve the growing container
throughput demand, yielding fruitful results in output and
performance of the port through the planning, developing
and building new ports, port facilities, infrastructures, and
terminals etc. Furthermore, the findings have also
contributed to closing the previously identified study gap.
Because the model and analysis on container throughput
has encouraged the export activities and domestic
productions and stimulated growth of economy, as well as
in order to propose and develop the maritime transport
strategic policy and development for Cambodia to support
the higher demand of ports throughput in the future.
Keywords:- Time Series, Econometrics, Vector Error-
Correction Model, Maritime Transport, Container
Throughput and Ports.
I. INTRODUCTION
According to Ng and Liu (2014), the container ports
have become the backbone of world logistics and supply chain
efficiencies, served as the crucial mode of world transport
services in terms of cost-saving, capacity benefits. Nam and
Song (2011) pointed out that sea trade through the ports
increased by nearly 90 percent in volume and roughly 70
percent in value of world trade. Scholars and industry practitioners have been drawn to study of container
throughput or performance measurement in ports and
terminals. According to Pallis et al. (2011) and Woo et al.
(2011a), the port and terminal performance could be
considered as a well-established part of the port-related
literature reviews; and Tongzon (2001); Cullinane et al.
(2004); Da Cruz et al. (2013) also demonstrated the container
throughput or port output conventionally focused on the
productivity and efficiency of port operation. With an
uncertain logistics environment, container ports have recently
faced with numerous challenges and restructuring in order to
continue and maintain. As a result, the up-to-date container ports are now parts integrated into complex systems that
operate in a volatile transportation characteristics.
This study aims to provide insights into the maritime
transport markets and ports, and to explain how important factors determine container throughput by empirically
analyzing and testing the equilibrium effects on long-run and
short-run relationships among key explanatory factors of ports
in Cambodia. Two major international ports are Phnom Penh
Autonomous Ports (PPAP), the inland port on the Tonlé Sap
River, accessed via the Mekong River, and Sihanoukville
Autonomous Port or Port Autonome du Sihanoukville (PAS)
as the seaport (PAS, 2015; PPAP, 2016). PPAP transfers
containers, vessels, barges, petroleum, and gas directly to
Phnom Penh. Sihanouk Ville Autonomous Port provides
access for other heavy goods that presently transported by road and rail. This paper also focuses on investigation and
estimation the demand for container throughput by using and
selecting the factors of determinants to see relation effects
whether there is a long-run and short-run equilibrium of port
performance to meet the supportive purposes that can give
policy implications for the port sector and maritime transport
and effective decision-making supporting tool to improve
efficiency and output.
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 376
According to PAS and PPAP, the total container
throughput of international ports in 2000 accounted for 130,435 TEUs, increased to 222,928 TEUs in 2010 and
459,839 TEUs in 2017, respectively. In 2016 and 2017, there
were about 400,187 TEUs and 459,839 TEUs going through
PAS, and showing a 15 percent increase over 2016. PPAP’s
container throughput achieved by 184,805 TEUs in 2017, up
from 62,256 TEUs in 2010 and 746 TEUs in 2002. PPAP
came to 184,805 TEUs, up 21.76 percent compared from
151,781 TEUs in 2016. Many previous studies and researches
have also examine and mainly focused on the forecasting the
port performance and efficiency and many others conducted
the container cargo volumes and flow of the container port
throughput. However, until today there has been no any study and research to determine the factors influenced on the
container throughput of Cambodia’s International
Autonomous Ports. For that reason, this research is intending
to fill the gap by attempting to investigate and examine to
identify and measure factors that determine the container
throughput of Cambodia's International Autonomous Ports
accordingly. This research question will mainly restrict to
factors determining maritime transport markets for
Cambodia’s ports by applying the time series econometric
analysis on the container throughput of Cambodia's ports. To
analysis this framework, the econometric analysis of VECM is proposed and used STATA software program to observe the
effect of the long-run relationships between container
throughput and four key factors determined, such as the
number of shipcall, export, import, and GDP per capita. In
terms of scope, this study does not specify any other related
area of economics and management in general due to the
limited constraint of time of study and data collection. This
research study may has many fruitful and advantageous
contributions for the country and the globe based on the
maritime economics and logistics and management as a
special instrument to promote the sustainable economic
growth and development of regulative principle and strategic policy. The maritime industries, governments, and port
authorities should follow and stay updated with the fluctuating
dynamics of the maritime transport sectors, as well as
movements and growths that affect it.
II. LITERATURE REVIEW
The literature review will conduct theoretical and
empirical overviews with reference to objectives and topics. The literature review first provides the theoretical overviews
of the maritime transport market and how the maritime
transport economics are related to broad trends in academic
research of maritime transport and ports. The review then
outlines academic research on analysis of port container
throughput, characteristics of changing port business
environment, and port output measurement to establish the
guideline of the conceptual framework model about port
output indicator selection as well as identifying the
relationship between the variables. The research review will
also present and investigate previous research findings of
other researchers to determine whether the study of Cambodia's port can be defined and identified through the use
of statistical and econometrical tools to measure and
investigate the long-run relationships of container throughput,
and to set out the maritime transport policy implications.
A. Theoretical Foundation
The research’s theoretical foundation and literature
reviews are selected the best theoretical and empirical
frameworks, hypothesis development and conceptual frameworkrelated to the key variables to reveal the research
gaps of study. Furthermore, the hypothesis development is
paramount related to the research’s objectives and presented
and discussed.
a) Port Performance
Many types of measures can be used to evaluate port
performance according to Langen, Nijdam, and Horst (2007).
The most commonly used indicator is the measurement of
goods throughput volume, which can be expressed and
measured in TEU or Ton, but another measurement of
monetary value can be used as well. The investigated study of
the influence of port supply, port demand, and value-added
activity of port on the development of regional economics in
five Chinese costal port clusters performed by Deng, Lu, and
Xiao (2013) showed that supply and demand for the port had a significant positive relationship with each other. Value-added
activity was found to be significantly related to port demand,
and value-added activity had a significant influence on the
regional economy, whereas port supply and port demand had
no direct significant impact on the regional economy. Port
supply, on the other hand, has had an indirect influence on the
regional economy through port demand and value-added
activity. Similarly, port demand influenced the regional
economy indirectly through value-added activities. United
Nations proposed several port performance indicators since
the seminal works (Unctad, 1976; De Monie, 1987). In recent
years, the studies of port performance and throughput has started to rise interestingly and to be analyzed in numerous
methods with utilizing small samples size of ports all over the
world, growing continuously (Chung, 1993 &Poitras et al.,
1996).
b) Port Throughput
Unctad (1976) and MRC (2018) define the meaning of
TEU (Twenty-Foot Equivalent Unit) as the standard
dimensions of container throughput, measures 20 foot (length)
8 foot (width) 9 foot, and is used to describe the carrying
capacity of a container ship, cargo ship and container port, can
load and unload onto the ships, truck, trains and planes.
Throughput is well-defined as a number of container activities
that occur as it passes through the port terminal (Liu and Park,
2011) as well as the amount of commitment required to
transfer cargos in terms of container movements per unit of time are measured. Container throughput, according to
Langen et al. (2007), is one of the most widely used
performance measures in the port industry. Alternatively,
various factors such as the number of ship call and export and
import or transshipment cargo, berth and yard utilization,
crane and ship productivity, and service time could have an
effect on port throughput growth. All of these are internal
factors that the port company or port authority can influence
directly or indirectly. Besides the internal environmental
factors that port authorities or port operators have direct or
indirect control over must be taken into account, it is also
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 377
important to consider external environmental factors based on
macroeconomics when evaluating the impact. When an output handle is assembled in a port, the port’s throughput is
determined by the performance of respective independent
industries (Paflioti et al., 2017).
B. Theoretical and Empirical Review Throughout many capturing and selecting literature
reviews, the study of many researchers dealt with, which just
merely or mainly compared a single port or many ports in
each country or in regional country as preferred, particularly
and practically in Hong Kong, Korea, China and ASEAN
countries etc. and many are also focus on the forecast of
container throughput. Yet there were not many articles or
researches focused on which factor has the strongest
determining or affecting to container throughput in Cambodia
and the main concern. The many attributes determined port
and the economic factors influence on attracting shipping liners and industries, and thus, throughput volume or
container throughput. The number of ship call, export, import,
and GDP per capita will be discussed as the four major
influences on the throughput of port.
a) Factor Affecting Container Throughput
The focus on this study is on the demand-side of
TEU’s output of Cambodia’s ports. From a practical
standpoint view, determining and predicting container
throughput by using the volume of throughputs in TEU rather
than the volume of cargo in Tonne produces numerous
advantages for ports, operators, and carriers (Levinson, 2006)
because operational decisions and services offered are directly
and indirectly dependent on the number of loading and
unloading changes per unit. Numerous studies examining the
factors affecting the maritime market and the factors affecting
the container port throughput has decreased significantly in recent years. To assist in analyzing container throughput
determinants, the impact on the port’s container handling
performance will need to be considered. The throughput in a
container can be measure and stored in many different
behaviors (Langen et al., 2007; Liu Bin et al., 2002; Jiang, J.
et al., 2007; ArjunMakhecha, 2015-2016). The utmost
commonly used indicator is the throughput volume of cargoes,
which is defined as the number of containers in TEU. An
increase in throughput is regarded as evidence of the output
efficiency with which container ports deal with throughput
(Langen et al., 2007). In this research, the author will present and examine the key selected determinants of container port
throughput founded by the related reviews.
b) Number of Ship Call
In literature reviews, Barros (2005); Tovar and Trujillo (2007); Gonzalez and Trujillo (2008) identified the number of
ship call as one of the multiple outputs of a port at a given
time, i.e., this study is conducted annually, as the number of
ship call. There are a number of perceptions regarding key
players in the decision making process for terminal or port
selection. Shipping lines are the most important players in
determining which port to use. Ports play an important role
when it comes to value-driven chain systems. Ports and their
services need to provide customers with consistent value
compared to their competitors in value chain systems. The
reality is that many industries, either shippers or shipping
lines, determine cargo flow, which will look for a route that
will provide the lowest price for a given level of servicewhile maintaining quality. Container ports capable of providing this
service will be selected as call ports in the logistics chain,
acting as a node in the chain (Yap and Lam, 2006). According
to Kutin (2017), after the annual throughput of containerized
goods as a general indicator, the number of ship call is another
possible indicator in measuring port’s performance. Several
researchers emphasized the number of shipcall, which refers
to the frequency with which shipping lines come and call at a
port with more competitive service and cargo flows (Lam and
Yap, 2008). In the analyses by Slack (1985); Bird (1988);
Bardi (1973); Brooks (1984 & 1985); Saleh and Lalonde
(1972), shipping lines are the key players in determining port choice, value-driven chain system, and port choice. According
to Unctad (2004), the shipping connectivity index by country
includes the number of ships, carrying capacities, and
services, as well as the number of shipping companies, ship
call, and maximum vessel size. In the calculation of PAS and
PPAP, the total number of ship call measured in the unit of the
ship also included a general cargo ship, an oil tanker, a
container ship, and a passenger ship.
The capability of a seaport to attract maximum container
ship traffics, such as ship call, container ship, and the
frequency of ship arrived at ports, are factors of container
throughput. Ship owners are increasingly selecting the port
choice to call (Huybrechts, 2002; Hilde Meersman, 2016).
Cullinane (2004); Brooks (1985) revealed that increasing the
frequency or number of ship call can attract more customers and cargoes, resulting in an increase in cargo volume and
throughput, which will have a positive and significant effect
on port performance (Song and Han, 2004). Moreover, an
increase in ship call frequencies is attracting both shippers and
shipping lines alike. Slack (1985) conducted a study that
revealed some intriguing information about port selection.
c) Import and Export
For the time being, as soon as it is aware that, not
many studies are published that review the factors affecting
container throughput volumes or port throughput handling, i.e.
the trade volume of container throughput – export and import
of container throughput. There are now only a few papers that
attempt to review a part of the literature on forecasting
container throughput within the multi-port region (Jiaweia et
al., 2012).By theoretical and empirical models and various numerical studies for the Asia-Europe and Asia-North
America trade lanes found that multi-purpose of ports was
superior in terms of the total cost. Theoretically, developing a
multi-port region in Asia trade is efficient, as evidenced by the
rapid container throughput growth. Thus, in terms of a multi-
port region, the analysis of port throughput in the North
America, Japan, and Southeast Asia and in the Pearl River
Delta is both necessary and feasible (Imai et al., 2009).
According to Drewry (2010), container throughput benefited
from the estimation of consumer price index in selected
developed and developing countries such as the United States,
the European Union, and BRIC countries such as China and India.
Therefore, these actors are required for the
operationalization of container throughput handling at ports;
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 378
alternatively, maritime transport is the derived demand for the
economic system.Moreover, when determining the container throughput, it is important to consider the macroeconomic
components of a country's trade value, such as export and
import, among other things. Tongzon (1995) accepted that,
these factors significantly impact container throughput
handling at ports. According to some studies, the value of
import and export into a country is the most important factors
(Seabrooke, W. et al., 2003), and it is also discovered a
nation's economic development is largely responsible for port
expansion. This container throughput considers both inter-port
interactions and other influences. Nonetheless, Jiaweia et al.
(2012) confirmed that a demand-based module provides
incremental data as well as the impact of random events on the maritime industries, which was required to forecast
container throughput volumes at ports in South China's Pearl
River region during the global financial crisis.
d) GDP Per capita
Langen et al. (2012) argued that selected factors, such
as growth of GDP and growth of trade are used to prove the
existing relationship among the throughput and
factors.Langen (2003) also developed a model for analyzing
and forecasting future demand for transportation between two
countries. Within the framework of reference of the
investigation, it emphasizes the high degree of uncertainty in
trade flows and attempts to explain the pivotal need for
flexibility in economic interest decisions relating to port
investments. Andrea Mainardi (2016) explored cargo
throughput forecasting in Portuguese ports using econometric indicators related to the Portuguese economy, such as
Portuguese GDP, population, inflation, domestic
consumption, and world GDP.
C. Econometric Review According to Liu and Park (2011); Seabooke et al. (2003);
Gosasang (2011); Peng and Chu (2009); Langen et al. (2012)
conducted research on forecasting demand for container
throughput in TEU by using the regression model. The effect
of container throughput on port performance will need to be
measured to aid in analyzing container throughput
determinants. Port performance can be measured in several
ways (ArjunMakhecha, 2015-2016). Most studies that used
the regression approach to analyze container throughput used
GDP as an independent factor. Containerized trade flows can
be estimated using long-term global GDP growth, and the multiplier effect of container growth results in three to four
times GDP growth (Unctad, 2013). Liu B. et al. (2002) also
used a linear regression model to examine the relationship
between container throughput and GDP, fixed asset
investment, interest rates, and international trade. Using per
capita GDP data, Nakano H.P. (2004) discovered that rising
per capita GDP in major developed countries would lead to a
decrease in the number of containers per capita. Jiang et al.
(2007) conducted a value-based economic analysis of
container throughput, local GDP, and international trade
volume using a binary linear regression model, and Xu Xiao
(2014) also examined the relationship between natural miniaturization of GDP and container throughput using a time
series method.
The econometrics model is the most useful method for
determining the relationship between variables. If the relationship is known between them, there can be a significant
change among the variables over the forecast limit
(Armstrong, 2001). Several techniques applied to determine
and estimate or predict container volume throughput, such as
Neural Network (Gosasang et al., 2011), Regression Model
(Chou et al., 2008), Grey Forecasting Model (Qiuhong, 2009),
and Vector Error Correction Model (VECM) (Rashed et al.,
2013), etc. Much research is found in container analyzing
forecasting that indicates a correlation between
macroeconomic variables and container throughput of ports
(Chou et al., 2008b). Thus, in this study, The VECM will be
used to investigate and analyze whether a long-run relationship exists between the container throughput of
Cambodia's combined international ports, namely PAS and
PPAP. In most literature reviews, many researchers have
implemented the VECM as the best model to examine and
estimate the container volume demand in the long run effect
in different ports. The empirical reviews of the econometric
approach of VECM conducted by many researchers and
scholars to estimate the container throughput are more
accurate and reasonable than regression analysis (Syafi’I,
2005).
D. Development of Hypothesis
This study can conceptualize framework analysis on
variables in the estimated model, a basic conceptual
framework or structure organized around the theory. It defines
the kind of variables that are going to be conducted in the estimation and analysis. Four exogenous variables, namely the
number of shipcall, export, import, and GDP per capita, are
categorized as factors determining the container throughput of
Cambodia's combined international ports. Container
Throughput measures in TEU, which is the endogenous or
dependent variable. This study aims to test the four
hypotheses which are previously mentioned and considered as
the claim or assumption about the value of a population
parameter. The proposed hypotheses are as the following;
H1: The number of ship call is statistically significant with
container throughput.
H2: Import is statistically significant with container
throughput.
H3: Export is statistically significant with container
throughput.
H4: GDP per capita is statistically significant with container
throughput.
III. METHODOLOGY
A. Research Design
The correct econometric methodology and research design
are found to be critical for using and testing the data. To
begin, the unrestricted Augmented Dickey-Fuller (ADF) and Phillips Perron (PP) tests are employed to examine the
stationarity of series among variables. Second, Vector
Autoregression (VAR) is utilized to select the most
appropriate lags. Third, the Johansen Cointegration test is
performed to see if there are any cointegrating vectors or
maximum ranks. The cointegration test determines long-term
cointegration amongst variables. Fourth, constraints are
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 379
placed in the model of VAR, and VECM is estimated to
determine effects of long-run effects and short-run dynamics. To maintain the current results and findings, Impulse
Response Function (IRF) is also used in this study.
Furthermore, model diagnostic tests such as system stability
tests, autocorrelation, heteroskedasticity, and normality tests
are run to ensure the model’s reliability and validity and to
verify the findings. The VECM, on the other hand, is
estimated using the STATA/IC15.0 software program.
Unknown parameters can be determined using a variety of
analysis techniques and estimates. Nonetheless, the VECM
model is much more efficient and accurate in its research and
detailing of long and short-run equilibrium relationships.
B. Data for Analysis
This research is utilized the quantitative approach from
1992 to 2017 to analyze and assessed in annual time series
data. Further, the dataset was transformed to logarithm to get elasticity coefficients. The data of container throughput of
Cambodia’s port was included import and export of laden and
empty container throughput in TEUs. Container throughput
data was collected from PAS and PPAP and aggregated and
combined into the total container throughputs of Cambodia’s
ports. In addition, container throughput variable is used as
dependent variable. The number ship call measured in unit of
ship was also collected from PAS and PPAP ports of
Cambodia, and included general cargo, oil tanker, container
and passenger ship. Export and import and GDP per capita
value measured in USD million which was collected from the
IMF. The data of annual GDP per capita was calculated by dividing annual GDP to total population of Cambodia from
1992 to 2017.
C. Econometric Model
The container throughput was taken as the dependent variable. The number of ship call, export, import, and GDP
per capita were chosen as the independent factors to explain
and control container throughput in Cambodian ports.
Multiple regression analysis was used to identify factors that
are related to the dependent variable and to investigate the
nature of the relationship as well as to conduct research,
prediction, and forecasting. Multiple regression models were
used to investigate causal relationships between independent
and dependent variables.
The equation formula and selected variables are written
as follows:
lncontt = β0+β1lnnsct+ β2expt+ β3impt + β4gdppct+ εt (1)
Where,
Container Throughput in TEU = cont
Number of Ship Call in Unit = nsc
Export in USD = exp
Import in USD = imp
GDP Per Capita in USD = gdppc
β0: The Intercept β1, β2, β3, and β4: Coefficients
t: Time Series
εt: Error-Term
All variables are basically transformed into natural
logarithms. Transforming the nature of dependent and
independent variables to logarithmic form help us to measure
the elasticity of dependent variable with respect to the independent variables. β0 is constant or intercept, and β1, β2,
β3, and β4 are the parameters or slopes measured by each
independent variable, and whereas the εt is the long run error-
term or random disturbance term. The time series econometric
model was used in this study to investigate cointegration
vectors and causality directions of demand for container
throughput of combined ports in Cambodia.
D. Step of Estimation
Step 1: Data must be stationary.
Step 2: Optimal lag is selected.
Step 3: Johansen cointegration test is identified.
Step 4: VECM is estimated.
Step 5: Diagnostic tests is performed.
E. Time Series and Stationary
The time series model uses the movements of variables in
the past to estimate values and results in the future. The
timeseries model, according to Stock and Watson (2003), is
not economic theory. However, unlike structural models, the
predictability of the estimated equation should be evaluated based on the model's performance outside the sample.The
timeseries econometric model, according to Keck, Raubold,
and Truppia (2009) can produce relatively accurate
forecasting in most cases, particularly when variables have
multi-dimensional relationships. Because of its complexity, it
is vulnerable to omitted variable bias, misspecification,
multiple simultaneous causalities, and other issues, all of
which can result in significant forecasting errors.According
Gujarati (2012) pointed out that the value of covariance
between two time series of period is determined only by the
distance or gap between two time series of periods, rather than the actual time series at which the covariance is computed,
indicating that a time series is stationary if its mean and
variance remain constant over series of time.
F. Unit Roots Test Unit root tests are used to determine whether or not data
sets are stationary. According to Gujarati (2008), the unit root
test is a stationarity or nonstationarity test that has received
widespread popularity over several years due to simplicity
(Gujarati, 2008). To evaluate if the series is stationary in level
I(0) or nonstationary in level, but stationary in first
differencing I(1), i.e. integrated of degree on I(I), the unit root
test of time series is carry out . Dickey-Fuller suggested that
ADF for unit root test (Engle and Granger, 1991) is conducted
with a constant and trend term based on the critical values.
The null hypothesis test of unit root is the only assumption that is rejected. To determine the appropriate or optimal
lagged selection to ensure that the error becomes
approximately white noise, the Akaike Information Criterion
(AIC) and the Schwarz Information Criterion (SIC) are used
for testing purposes. Standard tools for stationary testing
include the enhanced Dickey-Fuller and Phillip-Perron
methods. Because the PP test corrects for serial correlation,
autocorrelation and heteroskedasticity problem in the errors
disturbance, the estimation finds preferable to the ADF test
for many applications. Furthermore, PP tests do not
necessitate lag selection and are based on a serially linked
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 380
regression error term rather than a random effect model. ADF
tests are similar in that estimates are centered on the null hypothesis that the series is not stationary. The null for PP
tests are similar (Dickey F., 1979; Perron P., 1989).
G. Cointegration Test
Engle and Granger (1987) proposed that presence of number of cointegration vectors addressed the problem of
integrating the dynamic effect of short-run relationship with
long-run equilibrium. This approach, which has been
frequently utilised in quantitative studies but has a number of
drawbacks that include distortion in sample size and non-
unique characteristics depending on the normalization
variable employed, cannot identify numerous cointegrating
vectors (Banerjee et al., 1993). Engle and Yoo (1991);
Johansen (1991, 1998) proposed and suggested a three-step
approach and maximum likelihood model. There are
numerous cointegration tests, and cointegration may have when the test is performed. The most commonly used is the
multivariate test based on the autoregressive representation
developed by Johansen (1988 & 1991); Johansen, Juselius
(1990 & 1994); Gonzalo (1990 & 1994). Johansen and
Juselius (1988 & 1991) were the first to identify the
cointegrating vectors in a system, providing a maximum
likelihood technique to test for the existing presence of
cointegrating vectors in the system. This method, which is a
variation of the VAR method, is commonly referred to as
Johansen Maximum Likelihood or VECM in the scientific
literature reviews. In the VAR framework, all of the variables
in the VAR model are identified and evaluated utilizing theoretical economic theory. It needs to conduct a pretest of
the series for unit root before applying the VECM technique.
To identify the number of cointegration of Johansen process,
the Johansen maximum likelihood technique provides and
constructs two different types of likelihood ratio tests: the
trace test and the max eigenvalue test, which are both
employed. The Johansen technique has better properties than
other estimators (Johansen &Juselius, 1990, 1994; Gonzalo,
1994), and the finite sample properties are consistent with
asymptotic findings. The VECM can be analyzed by using the
Johansen technique, which includes the error correction term or speed of adjustment in each equation. When the computed
trace test and max eigenvalue test statistics are greater than
the critical values of 1 percent, 5 percent, or 10 percent, the
null hypothesis of no cointegration is not accepted. If the
estimates of trace and eigenvalue statistics are less than the
critical value, the alternative hypothesis of one or more
cointegrating vectors is accepted.
H. Vector Error-Correction Model (VECM)
The VAR equation is one of the particular variants of the
system simultaneous equation. The VAR model can be
employed if all of the variables are stationary. When the
variables in the vectors, on the other hand, are not stationary,
then the model used is the VECM if the variables in the
vectors have at least one or more cointegration relationships.
Enders (20150) defined VECM as a VAR that built
specifically for use with nonstationary data that has a cointegrating relationship with the original data. The time
series model is one of the few that can accurately analyze the
level at which a variable can be restored to equilibrium
following a shock to other variables. The VECM method is
extremely beneficial for estimating the short-run and long-run effects in series data. Under the assumption of cointegration,
the VAR can be represented as VECM, as shown below.
∆contt = α+ ψcontt−1 + ∑ Гi∆𝑐𝑜𝑛𝑡t−i + θZt−1k−1i=1 + εt (2)
(2)
α = The Intercept
ψ = Vector of Intercept (αβ’)
k-1 = Lag Length
Г = Vector of Variables [cont, nsc, exp, imp, gdppc]
θ = Speed of Adjustment
εt = Residual or Error-Term Zt-1 = Error-Correction Term (ECT)(-1)
contt = Vector of Lagged Variable
β’contt, Δcontt = Coefficient of Vectors
Utilizing an econometric model is a clear framework for
developing and testing economic hypotheses. The model
demonstrates anfull insights of the effect and dynamic
behavior. The use of likelihood methods providesfor
undertaking inference without the need to derive estimator
and tests.
Unrestricted VAR is the estimate of ψ matrix applied and
tested by using the Johansen's approach whether the restricted
suggested by the reduced rank of the matrix is rejected. This
equation can be used to indicate the rank N x N matrix as ψ =
α x β’ where N x R matrix of rank. According to the model
employed in this investigation, N = 5, cont = (cont, nsc, exp,
imp and gdppc). b’ contt is stationary and is viewed as long-
run equilibrium relationship between the jointly influenced
factors because b’ is a matrix that indicating the cointegration
relations, It is critical to highlight that unless a normalization or identification procedure is specified, it is impossible to
predict the individual coefficient b’. In the short run,
stochastic shocks may push and drive the system, but in the
presence of a cointegration relationship, pushing variables that
will lead the system to intersect with the long-run
relationship.The stationary process proved byb’contt, and the
speed of adjustment coefficient formed by A for the equation
showed the equilibrium relation and deviation. The max
eigenvalue and the trace test are utilized to determine the
number of cointegration vectors under the ψmatrix of reduced
rank hypothesis. Ifψ has a rank of zero, no stationary linear integration can be found, indicating that the variables in contt
are not cointegrated. If rank R of ψ matrix > 0, then R is not
possible nonstationary linear integrations after considering the
parametric constraints implied by long-run relationships, the
short-run function is estimated steadily in consistent manner.
In response to short-run disequilibrium, the VECM helps
numerous factors that influence simultaneous dynamic
adjustments at different rates than the others because theory is
frequently unsatisfactory for defining the dynamic adjustment
process. By presenting a multivariate VECM, this study aims
to investigate how the key elements of independent variables affect container throughput.
I. Long-Run Causality
Johansen and Juselius (1990) developed the economic
relationship of long-run equilibrium effect. The Johansen
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 381
cointegration test typically comprised of 3 general steps,
which are as follows: First, ensure that all variables in the model are integrated in the same order by running unit root
tests on each variable. Second, calculate the appropriate lag
selection for the VAR model in order to ensure that the
estimated residuals are not autocorrelatedwith the observed
residuals. Three, Enders (2004 & 2010) proposed to estimate
VAR model in order to generate the cointegration vectors,
which identify the rank of cointegration using the trace test
and the max-eigenvalue statistical tests. The coefficient of
error-correction term (ETC) of one-period lagged value
indicates the causality of long-run effect and expresses the
adjustment speed. The expected value of 𝛽t should be
significant and have a negative sign. 𝛽t (-1&0) of absolute
value indicates how quickly the equilibrium effect is
reinstated and restored. The significance of ECT implies that
the long-run equilibrium relationship of the independent
variable is moving and can explain the effect on dependent
variable.
J. Short-Run Causality
The jointly significant variables describe how the
endogenous variables respond to short-run dynamics. This
study, as discussed by Hakim and Merkert (2016), also
assesses the jointly significant level of the coefficients of the
endogenous variables 𝛾t and 𝛿t by testing the short-run
relationships effect.
K. Impulse Response Function
It is intriguing to determine the shocking reaction by
performing the Impulse Response Function (IRF) in one
variable to an impulse in another variable in a system with several other variables. It is said to have the latter occurrence
of the causally responsible for the earlier occurrence if there is
a shocking reaction in one factor of variable to an impulse in
another factor. The IRF takes a look for the moving average
reaction and explains how the factor influences a shocking
reaction over time to a single response growth in itself or any
other variables influences a variable over time to a single
response growth in itself. According to Lutkepohl’s (1991) research, the impulse shock or effect of dynamic multipliers
can be acquired using a K-dimensional VAR model with an
infinite moving average factor. The impulse shocking is an
effective and efficient measure for determining and a useful
tool in assessing the stability of the estimated equation within
the framework of the vector autoregression model. The
combined effect of shocking reactions and response to zero
indicates that a system is stability.
L. Post Diagnostic Check
The statistical diagnostic tests such astest for stability,test
for multivariate heteroskedasticity, and LM test of
autocorrelation and testing for normality was performed and
carried out on the VECM model to verify that it represents the
data generating process adequately and the appropriated
model.
IV. RESULT AND DISCUSSION
A. Descriptive Statistics
The econometric analysis could be guided by the nature of
available time series data and patterns of relationship between
container throughput and number of ship call, export, import,
and GDP per capita. The findings of descriptive analyses may
also provide better hints and insights about the causal
relationships of the variables discussed in the following sections. The growth of container throughput at Cambodia's
combined ports has been studied in relation to the number of
ship call, export, import, and GDP per capita. For comparison
purposes, the ports that have encouraged and enabled
Cambodia's maritime transport market are also taken into
account. Correlation coefficients could present the
relationship between two variables.
Summary Statistics of the Variables (1992 to 2017)
Variable Obs Mean Std. Dev Min Max Skewness Kurtosis
lncont 26 11.98846 1.02611 9.96 13.38 -0.66849 2.2884
lnnsc 26 7.578077 0.365885 6.84 8.12 -0.45156 2.4558
lnexp 26 8.297308 1.066469 6.55 9.97 0.08363 1.630
lnimp 26 7.781538 1.377768 5.22 9.7 -0.36930 1.9906
lngdppc 26 6.231538 0.5829147 5.49 7.23 0.3128 1.5741
Table 1: Descriptive Results
Source: Author’s calculation
It has been noted that lncont has the highest average of
11.99 percent, while lngdppc has the lowest average of 6.23
percent. According to table 1, the lnimp is the most volatile at
1.378 percent, ranging from 5.22 percent to 9.7 percent,
whereas the lnnsc is the least volatile at 0.36 percent. Table 1
also demonstrates that lncont, lnimp, and lnsc are negatively
skewed, whereas lnexp and lngdppc are positively
skewed. A normal distribution curve's kurtosis measures the
peak in the center of the curve. The distribution is described
as mesokurtic, which implies that it is normal, if the value is close to three. The distribution is leptokurtic as long as the
number is greater than or equal to 3. As a result, all of the
series with values less than 3 in table 1 exhibit platykurtic
behavior (Wooldridge, 2010).
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 382
Correlation Coefficients Between Variables
Variable lncont lnnsc lnexp lnimp lngdppc
lncont 1
lnnsc 0.9684 1 lnexp 0.9326 0.9546 1
lnimp 0.9834 0.9716 0.965 1
lngdppc 0.8805 0.9185 0.9857 0.9236 1
Table 2: Correlation Coefficients
Correlation Coefficients Between Variables
Variable lncont_d1 lnnsc_d1 lnexp_d1 lnimp_d1 lngdppc_d1
lncont_d1 1
lnnsc_d1 0.3871 1
lnexp_d1 0.3497 0.3228 1
lnimp_d1 0.2108 0.1127 0.0972 1
lngdppc_d1 0.1832 0.3132 0.4968 -0.1564 1
Table 3: Correlation Coefficients
Source: Author’s calculation
Before establishing the VECM and avoiding a multi
collinearity problem, the correlation analysis of all factor
variables will be performed. If there is a correlation between
the independent variables and they have an approach to
influence the dependent variable. The correlation matrix of
two-sided Pearson Correlations between all variables is shown
in tables 3 and 4. Correlations show that all variables are related to one another. Number of ship call, export, import,
and GDP per capita all has significance and strong
relationships with container throughput. Table 2 shows the
coefficient correlation for the entire series. The results show
that the model's dependent variables have a strong positive
correlation relationship. Because each dependent variable is
highly correlated, their presence in the same model will result
in a multicollinearity problem. As a result, it is decided to
look into their impact on the first difference. Similarly, the
results of the variables at first difference presented in Table 3
show a positive relationship, with the exception of the correlation coefficients between lnimp_d1 and lngdppc_d1.
The covariance matrix in tables 4 and 5 shows that the
coefficients of all variables are positive, indicating that as one
variable increases, so does another. But except for the
negatively related coefficient between lngdppc_d1 and
lnimp_d1. Descriptive analysis and graphical representative
indicate that there is most likely a long-run and short-run
relationship between the study's dependent vs. independent
variables.
Covariance Matrix between Variables
Variable lncont lnnsc lnexp lnimp lngdppc
lncont 1.0529 lnnsc 0.363581 0.133872
lnexp 1.02054 0.372487 1.13736
lnimp 1.39023 0.489775 1.41786 1.89825
lngdppc 0.526687 0.195895 0.612784 0.741801 0.33979
Table 4: Covariance Matrix
Covariance Matrix between Variables
Variable lncont_d1 lnnsc_d1 lnexp_d1 lnimp_d1 lngdppc_d1
lncont_d1 0.024689
lnnsc_d1 0.004982 0.006708
lnexp_d1 0.007393 0.003557 0.018098
lnimp_d1 0.00558 0.001555 0.002201 0.028374
lngdppc_d1 0.002082 0.001856 0.004835 -0.00191 0.005233
Table 5: Covariance Matrix
Source: Author’s calculation
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 383
B. Stationarity Check
Before selecting an appropriate model, performing estimation, and avoiding erroneous analysis and
spurious results, it is necessary to determine whether each
variable is stationary or nonstationary. The purpose of this
analysis is to see if the data has stationarity properties using
graphical analysis, Dickey-Fuller, and Phillips-Perron tests.
Table 10 shows the results of two different tests that were
carried out. Both test revealed that the series levels
under discussion are I(1), that is, not stationary. Furthermore,
the graph of autocorrelation and the partial autocorrelation
function (ACF) are shown. Correlogram techniques are also
used to ensure the accuracy of the results.
The VECM model technique is being used to determine
whether there is a relationship between container throughput
and the number of ship call, export, import, and GDP per
capita. It is also interesting in determining a long-run effect and equilibrium. Before estimating the model, the statistical
conditions of series must be checked to determine whether
they are stationary or not. The graphs of the series under
consideration are depicted in Figures 1 and 2. If the time series is not stationary, the series may be typically converted
to stationarity using a technique known as differencing.
Because the variables in Figure 1 are non-stationary, so it is
best thing to turn them into stationary variables. To be valid,
the VECM model requires that the time series must
be stationary. The stationary variables after first differencing
are shown in Figure 2. The graph depicts a constant mean and
variance over time. It is necessary to conduct Q-statistic tests
to determine whether the variables are correlated to assess the
absence of autocorrelation. Correlograms of all variables at
the same significance level show that the relationship is
statistically and significantly related. The Q-statistic test is used to determine whether or not there is an autocorrelation
issue. Table 6 displays the Correlogram Q-statistics test of
lncont.
Fig. 1: Trends of All Variables Fig. 2: Trend of All Variables at I (1)
68
10
12
14
1990 2000 2010 2020year
lncont lnnsc
lnexp lnimp
lngdppc
-.2
0.2
.4.6
1990 2000 2010 2020year
lncont_d1 lnnsc_d1
lnexp_d1 lnimp_d1
lngdppc_d1
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 384
Table 6: Results of Correlogram lncont
Table 7: Results of Correlogram lncont_d1
Source: Author’s calculation
Autocorrelation Coefficient (AC) is close to zero,
indicating that the autocorrelation problem has been fixed.
The Q-statistic test result specifies that, despite being
statistically significant, Partial Autocorrelation Coefficients
(PAC) for all other lags is not statistically significant. Table 7
shows that the correlation Q-statistic test lncont_d1 is not statistically significant at the 0.05 of critical level, indicating
that there is no significant serial correlation in the model. As a
result, the test estimates imply that there is no evidence of
autocorrelation. The results show that the observed Chi-square
probability value is not statistically significant for critical
values of 5 percent or less. It is concluded that the estimated
model contains no presence of serial correlation. In this regard,
the model is appropriate for economic analysis. The graph
demonstrates that AC and PAC are significant at the level.
Because the p-values of the Q-statistic test are greater than 5 percent after converting the time series data to I(1), the null
hypothesis for all variables can be accepted. As a result, each
variable confirms to be stationary. Figure 4 shows that there is
no evidence of autocorrelation in the white noise.
11 -0.1094 0.2616 65.779 0.0000
10 -0.0480 0.1649 65.198 0.0000
9 0.0186 0.1902 65.093 0.0000
8 0.0762 0.1008 65.079 0.0000
7 0.1601 0.2326 64.844 0.0000
6 0.2574 0.0419 63.862 0.0000
5 0.3617 -0.2132 61.451 0.0000
4 0.4652 -0.0296 56.915 0.0000
3 0.5849 0.3124 49.754 0.0000
2 0.7402 -0.0450 38.926 0.0000
1 0.8753 0.9521 22.309 0.0000
LAG AC PAC Q Prob>Q [Autocorrelation] [Partial Autocor]
-1 0 1 -1 0 1
10 -0.1553 -0.1287 7.708 0.6573
9 -0.1484 -0.0929 6.6228 0.6763
8 -0.0182 -0.0989 5.6934 0.6815
7 -0.0028 0.0563 5.6802 0.5775
6 -0.1719 -0.1503 5.6799 0.4600
5 0.0722 0.0122 4.6304 0.4626
4 0.2808 0.2711 4.4543 0.3480
3 0.0382 0.1453 1.9197 0.5892
2 -0.2270 -0.2437 1.8748 0.3916
1 0.1135 0.1140 .36254 0.5471
LAG AC PAC Q Prob>Q [Autocorrelation] [Partial Autocor]
-1 0 1 -1 0 1
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 385
Fig. 3 : Autocorrelation at I(1) Fig. 4 : Autocorrelation at Level
Source: Author’s calculation
C. Unit Root Test
The logarithmic form of the original series were utilized to
solve and eliminate the possible problem of heteroskedasticity and to make the series more comparable between themselves.
The Phillips-Perron test of stationary similarly reveals that all
level series have unit roots; however, first differencing series
are determined to be stationary at least at the 5 percent
significant level. To ensure robust results, the Dickey-Fuller
and Phillips-Perron tests are carried out by applying a constant
and trend term to the test. Because the null hypothesis of unit
root is not accepted at I(1), and the variables become
stationary at I(1), which is also known as integrated order 1 or
I(1) (1). The Dickey-Fuller and Phillips-Perron tests are
analyzed to validate the confirmation of results. The estimated
outcome demonstrated that all variables uncovered are nonstationary at the level. Because the null hypothesis is not
accepted, it turned out that these variables become stationary
at I (1). Tables 10 shows that the augmented Dickey-Fuller
and Phillips-Perron models used to evaluate the data are
stationary. The Dickey-Fuller and Phillips-Perron tests have a
critical value of -4.380 at a 1 percent significance level, -3.600
at a 5 percent significance level, and -3.240 at a 10 percent
significance level. For the 5 percent, the nonstationary null
hypothesis is rejected.
Unit Root Test
Variables
Augment Dickey Fuller Phillips-Perron
Order Test Statistic Test Statistic
Level First Difference Level First Difference
lncont -1.379 -4.582 -1.309 -4.648 I(1)
lnnsc -2.801 -5.299 -2.680 -5.375 I(1)
lnexp -2.303 -5.571 -2.416 -5.566 I(1)
lnimp -1.907 -6.933 -1.872 -6.699 I(1)
lngdppc -1.773 -3.765 -1.853 -3.810 I(1)
Table 10: Dickey-Fuller and Phillips-Perron Tests
Source: Author’s calculation
D. Lag Length Selection
The long-run relationship effect is estimated using the Johansen cointegrating technique, which tests for the
significant existent presence of the long-run relationship
effect.However, before running this test, which employs a
VAR model, it is necessary to ensure that the VAR's lag
specification is correct. The estimated result of this test is
positive, as shown in table 11. The analysis results show that
all variables are nonstationary at level but stationary in I(1)
after differencing. However, before performing this test,
which makes use of a VAR model, it is necessary to ensure
that the VAR's lag specification is valid. As stated in table 11,
the estimated result of this test is positive. The analysis result shows that all variables are nonstationary but stationary in I(1)
after differencing at level. It is possible to determine whether
or not these series are cointegrated in the long run equilibrium
by consistently integrating them across time. To estimate the
long-run effect between the variables, the appropriate
lagged order of the VAR model must be determined using
lagged selection criteria. Because of the series' small sample
size, the 2 lag selection will be applied by auto lag selection.
-0.5
00
.00
0.5
0A
uto
co
rre
latio
ns o
f ln
co
nt_
d1
0 2 4 6 8 10Lag
Bartlett's formula for MA(q) 95% confidence bands
-1.0
0-0
.50
0.0
00
.50
1.0
0A
uto
co
rre
latio
ns o
f ln
co
nt
0 5 10Lag
Bartlett's formula for MA(q) 95% confidence bands
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 386
Table 11: Appropriate lag selection
Source: Author’s calculation
E. Cointegration Test
The Johansen technique is used to determine the existence
and number of cointegrating vector relationships, which is
used to calculate the tests of max eigenvalue (max) and trace
statistics. The null hypothesis of absent cointegration, that is,
no cointegration (Rank = 0), is not accepted by the max and
trace test statistic at a 1 percent or 5 percent significance level. As a result, the null hypothesis is not counted by the max and
trace test results. On the contrary, it is not rejected the null
hypothesis in these criteria where r ≤ 1,2,3,4 compared to the
alternative hypothesis with the value of r = 2, 3, 4 at 1 percent
significance level. Table 12 shows that the λmax test result of
44.6275 is not less than the critical value of 33.46 and 38.77 at
0.05 and 1 percent significance level, thus revealing the null
hypothesis of no cointegration is not accepted, and there is at
least one cointegrating vectors of equation. Table 12 further
supports this conclusion. The calculated trace test of 96.1011
is also greater than the critical values of 68.52 and 76.07 at
level 1 percent or 5 percent that is the evidence of
cointegration in the series. The λmax test result of 51.4737 is
still more than the critical value of 47.21. However, less than
the critical value of 54.46 at 1 percent significance level.
However, the results show that the λmax 25.4173 is not
greater than the critical values of 27.07 and 32.24 at 1 percent or 5 percent, respectively. Thus, the null hypothesis is not
rejected, and one cointegrating equation cannot be rejected.
The value of optimal lag length and the optimal order of
vector autoregressive model are chosen 2 shown in Table 11.
By Johansen's approach result, it is found that there are one
cointegration equations shown at both 5 percent and 1 percent
level at the Max-Eigen value or Max statistic value result, and
also one cointegration equations at the same level of 1 percent
in the trace results.
Johansen Tests for Cointegration
Rank Parms LL Eigen Value Trace Test Max Test 1% Critical Value 5% Critical Value
0 30 125.72347 . 96.1011 44.6275 76.07 68.52
1 39 148.0372 0.84425 51.4737*1 25.4173*1 54.46 47.21
2 46 160.74584 0.65322 26.0564**5 14.3436 35.65 29.68
3 51 167.91766 0.44990 11.7128 9.6180 20.04 15.41
4 54 172.72665 0.33018 2.0948 2.0948 6.65 3.76
5 55 173.77404 0.08358 - 44.6275 - -
Table 12: Trace and Max Statistic Test
Source: Author’s calculation
As a result, this study concludes this study opines that
one identified cointegration relationships are measured in the
Johansen approach among the variables at the 1 percent
significance level. This connection among variables specifies
the significant identification of a long-run economic
relationship. This link among the variables indicates the
significant identification of a long-run relationship. Finally,
only one cointegration vector equation's null hypothesis can be accepted. As a result, the cointegration test indicates that
the VECM, rather than a VAR, is appropriate for analyzing
the relationship among factors such as container throughput,
number of ship call, export, import, and GDP per capita
because the series are well-cointegrated.
F. Vector Error-Correction Model (VECM)
Therefore, this paper has applied this model to estimate the
long-term container throughput as combined international
ports of PAS and PPAP in Cambodia. The cointegration result
by the Johansen approach reveals that the influenced factors
of variables are cointegrated and have a long-run equilibrium
relationship. The calibration of the VECM is carried out using
the Johansen procedure for the estimation period from 1992 to 2017 with the time series data. VECM estimation result is
presented in Table 14, 15 and 16 as below. The number of
cointegrating equation among container throughput, number
of ship call, export, import, and GDP per capita in the
analyzed model is determined, and the estimation of VECM
parameters of multivariate time series is carried out, and the
output of the estimation is provided and seen that equation is
statistically significant and the model fits well.
Exogenous: _cons
Endogenous: lncont lnnsc lnexp lnimp lngdppc
2 173.774 75.96* 25 0.000 5.0e-11* -9.89784* -9.1816* -7.19813
1 135.794 208.31 25 0.000 1.1e-10 -8.81619 -8.42552 -7.34362*
0 31.6382 7.5e-08 -2.21985 -2.15473 -1.97442
lag LL LR df p FPE AIC HQIC SBIC
Sample: 1994 - 2017 Number of obs = 24
Selection-order criteria
. varsoc lncont lnnsc lnexp lnimp lngdppc, maxlag(2)
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 387
In Table 14, the results of the VECM model with all
variables such as container throughput, number of ship call, export, import, and GDP per capita are found to be
statistically and strongly significant at the 1 percent level of
significance, indicating a strong effect among all variables in
the long run equilibrium relationships. As a result of the
cointegration test establishing that the data is cointegrated, it
is now appropriate to estimate the VECM. Table 15 displays
the outcome of one cointegration vectorfor Cambodia's ports
in identifying the long-run relationships among variables.
The long-run equilibrium relations in the VECM are
represented in table 16 by the coefficients of the one rank of
cointegrating equation and trend and constant term. In the
long run, coefficients on lnexp and lnnsc have strong positive
effects on lncont, whereas coefficients on lnimp and lngdppc
have strong negative effects on lncont. The estimates of the
one cointegrating coefficient in the long-run equilibrium relationship (Johansen normalized on the coefficient) can be
interpreted as indicating that all coefficient parameters are
statistically significant at the 1 percent and 5 percent level.
The estimated coefficient of error-correction term (ECT) or speed of adjustment is negative sign of the coefficient on
_ce1 L1 value - 0.312 (error term from the first cointegrating
and lagged one-period) and strong significant at the 1 percent
and 5 percent level, which is the correct sign (negative) for
equilibrium and the probability of p-value is significant at
the 1 percent and 5 percent level. It is confirmed that there
is long-run economic equilibrium between dependent and
independent variables exist over time. The long-run
equilibrium causality is moving and running from lnnsc,
lnexp, lnimp, and lngdppc to lncont towards the equilibrium.
This implied that the ECM is well specified, well-identified.
This confirmed the expectation of a dynamic
adjustment toward long-run economic equilibrium relations.
Thus, if the error term _ce1 is negative in period t-1, which
can be seen as the demand for container throughput of PAS and PPAP being too high in comparison to the equilibrium
relationship other four determinants, then d cont must rise,
which can happen if the coefficient for _ce1 is negative and
significant. The greater negative coefficient on _ce1 L1, the
faster the correction.
To begin with, lnnsc and lnexp have a significant
positive effect on lncont. In other words, it suggests that, in the long run, an increase in demand for Cambodia's port's
container throughput is associated with an increase in the
number of shipcall that travelled (arrival and departure from
ports) and an increase in the export of cargo that port handled.
Second, lncont is significantly negatively affected by lnimp
and lngdppc. In the long run, it proves that the increase in
container throughput at Cambodia's ports is explained by the
deduction of imports and GDP per capita, and vice versa. This
can be expressed as an equation as follows;
lncont = –16.34352 + 1.82213lnnsc + 1.00068lnexp –
1.51536lnimp – 1.00532lngdppc (3)
The estimate result can be interpreted as long-run
elasticities. A 1 percent increase in the number of shipcall is
expected to increase container throughput by 1.82 percent,
and this estimate is significant at the 1 percent level. When
export rises by 1 percent, container throughput rises by 1
percent in response. Import is found to be negative to
container throughput in the long run equilibrium relationship.
A 1percent decrease in imports leads to a 1.5 percent increase in container throughput; this coefficient is significant at the 1
percent significance level. Similarly, a 1 percent decrease in
GDP per capita will have an increased impact on container
throughput of nearly 1 percent as well.
Following that, the estimates of the adjustment
parameters in table 17 for one cointegration equation, lncont,
lnnsc, lnexp, and GDP per capita are significant at the 5
percent level, except for lnimp. From the perspective of
empirical economic interpretation, the question is whether
imports adjust when the cointegration equation is out of long-
run equilibrium or disequilibrium.
α̂ (-0.3125, -0.2082, -0.2329, 0.1574,-01142)
β̂ (1.0000, 1.8221, 1.0006, -1.5153, -1.0053)
V̂ (0.0565, 0.0139, 0.1331, 0.3689, 0.0567)
Г̂
(
0.3805 0.7291 − 0.3496 − 0.2390 − 0.11520.2148 0.0074 0.0539 − 0.2558 − 0.1334−0.0963 − 0.1429 − 0.1416 − 0.2341 0.2146−0.1010 0.0765 0.0455 − 0.297 − 1.4128−0.0696 − 0.0541 0.1821 − 0.1904 0.0927 )
Table 14: Vector Error-Correction Model
D_lngdppc 7 .056013 0.7774 59.36823 0.0000
D_lnimp 7 .128344 0.7872 62.87356 0.0000
D_lnexp 7 .118232 0.7186 43.41968 0.0000
D_lnnsc 7 .073772 0.5897 24.43424 0.0010
D_lncont 7 .151375 0.6201 27.75378 0.0002
Equation Parms RMSE R-sq chi2 P>chi2
Det(Sigma_ml) = 3.02e-12 SBIC = -7.172096
Log likelihood = 148.0372 HQIC = -8.578559
AIC = -9.086434
Sample: 1994 - 2017 Number of obs = 24
Vector error-correction model
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 388
Table 15: Johansen Normalization Restriction Imposed
Source: Author’s calculation
Table 16: VECM
.
_cons -16.34352 . . . . .
lngdppc -1.005326 .5167297 -1.95 0.052 -2.018098 .0074452
lnimp -1.515368 .15391 -9.85 0.000 -1.817026 -1.21371
lnexp 1.000689 .402085 2.49 0.013 .2126167 1.788761
lnnsc 1.822134 .4472684 4.07 0.000 .9455042 2.698764
lncont 1 . . . . .
_ce1
beta Coef. Std. Err. z P>|z| [95% Conf. Interval]
Johansen normalization restriction imposed
Identification: beta is exactly identified
_ce1 4 835.2753 0.0000
Equation Parms chi2 P>chi2
Cointegrating equations
D_lnnsc
_cons .0565418 .066212 0.85 0.393 -.0732312 .1863149
LD. -.115222 .5276947 -0.22 0.827 -1.149485 .9190407
lngdppc
LD. -.2390774 .2299161 -1.04 0.298 -.6897048 .2115499
lnimp
LD. -.3496832 .2947484 -1.19 0.235 -.9273795 .228013
lnexp
LD. .7291129 .4489831 1.62 0.104 -.1508778 1.609104
lnnsc
LD. .3805853 .2570005 1.48 0.139 -.1231265 .8842971
lncont
L1. -.3125449 .1304574 -2.40 0.017 -.5682367 -.0568531
_ce1
D_lncont
Coef. Std. Err. z P>|z| [95% Conf. Interval]
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 389
Table 17: Adjusted Long-Run Relationships
Source: Author’s calculation
Table 18 : Short-Run Causality
Source: Author’s calculation
Overall, the results show that the equation model fits
well. The coefficients on lncont, lnnsc, lnexp, and lngdppc in
the cointegrating equation are statistically significant. The
ECT, or speed of adjustment toward equilibrium of lncont, has
a negative sign and is statistically significant (ce1 L1. = -0.31
and p-value = 0.017), confirming the existence of adjustment
toward equilibrium and identifying the existence of a long-run causality running from lnnsc, lnexp, lnimp, and lngdppc to
lncont.The error-correction components in the VECM
represent the speed of adjustment to the equilibrium.
Typically, the ECT should be between 0 and 1 and must have
a negative sign in value (Brooks, 2008). Table 18 shows the
estimates, standard errors, z statistics, and confidence
intervals for the short-run parameters. The parameters in this
model's adjustment matrix are the five coefficients on L. ce1.
The L. ce1 of D_lncont, D_lnnsc, D_lnexp, and
D_lngdppc is significant, indicating that the adjustment to
equilibrium is relatively slow. Table 18 shows the statistical
procedure used to test for short-run causality between these
variables. The null hypothesis is zero, which means it cannot
L1. -.114285 .0482728 -2.37 0.018 -.208898 -.019672
_ce1
D_lngdppc
L1. .157439 .1106095 1.42 0.155 -.0593517 .3742297
_ce1
D_lnimp
L1. -.2329378 .1018948 -2.29 0.022 -.432648 -.0332276
_ce1
D_lnexp
L1. -.2082494 .0635784 -3.28 0.001 -.3328608 -.0836381
_ce1
D_lnnsc
L1. -.3125449 .1304574 -2.40 0.017 -.5682367 -.0568531
_ce1
D_lncont
alpha Coef. Std. Err. z P>|z| [95% Conf. Interval]
D_lngdppc 1 5.604963 0.0179
D_lnimp 1 2.026001 0.1546
D_lnexp 1 5.226074 0.0223
D_lnnsc 1 10.72874 0.0011
D_lncont 1 5.739677 0.0166
Equation Parms chi2 P>chi2
Adjustment parameters
Prob > chi2 = 0.3397
chi2( 4) = 4.52
( 4) [D_lncont]LD.lngdppc = 0
( 3) [D_lncont]LD.lnimp = 0
( 2) [D_lncont]LD.lnexp = 0
( 1) [D_lncont]LD.lnnsc = 0
. test ([D_lncont]: LD.lnnsc LD.lnexp LD.lnimp LD.lngdppc)
Prob > chi2 = 0.2956
chi2( 5) = 6.11
( 5) [D_lncont]LD.lngdppc = 0
( 4) [D_lncont]LD.lnimp = 0
( 3) [D_lncont]LD.lnexp = 0
( 2) [D_lncont]LD.lnnsc = 0
( 1) [D_lncont]LD.lncont = 0
. test ([D_lncont]: LD.lncont LD.lnnsc LD.lnexp LD.lnimp LD.lngdppc)
Prob > chi2 = 0.0166
chi2( 1) = 5.74
( 1) [D_lncont]L._ce1 = 0
. test ([D_lncont]: L._ce1)
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 390
explain the lncont. The chi2 (5) value is 6.11, and Prob> chi2 =
0.2956; that is the p-value is greater than 5 percent, so the null hypothesis cannot be rejected and it implies that there is no
short-run causality running from lnnsc, lnexp, lnimp, and
lngdppc to lncont.
G. Impulse Response Function The impulse responses of each variable to itself are
determined, and other variables die out after a specific
duration. A container throughput shock responds positively to
itself and other variables. The early response of container
throughput to a unit shock in ship call, export, import, and
GDP per capita is positive and considered as significant. It is
proven that the container throughput is caused by the number
of ship call, export, import, and GDP per capita Granger-
Cause. Rising container throughput may be viewed as having
a significant impact on the number of ship call, export, import,
and GDP per capita. The outcome shows that among the
variables, the impulse response of shock exists to each other and becomes extinct after a specific period, confirming the
constancy of the estimate model. The impulse response of
number of shipcall to container throughput, the impulse
response of export to container throughput, and the impulse
response of import to container throughput are observed, and
it is observed that the number of shipcall and value of export
and import enacts clearly than the result frequencies of ship,
schedule of shipping line of shipcall when increasing, GDP,
and population, which also denotes the increasing volume of
throughout may have significant impacts. Additionally, it is
observed that the impulse response of GDP per capita to container throughput is a shock of GDP to container
throughput that gives a clear response. Furthermore, the
impulse response of GDP per capita to container throughput is
a shock of GDP to container throughput that produces a clear
response.
After that, the initial shock response of container
throughput to export and GDP per capita is positive and can
be considered significant. It is simple to illustrate that
increasing export will increase GDP per capita, thereby
increasing total container throughput. Figure 4 also asserted
that an import shock responds negatively to itself and all other
variables, whereas container throughput responds more
positively than other variables. It suggests that increasing
import will have a greater impact on increasing container
throughput than other variables. As shown in Table 16 and 17,
the equilibrium adjustment process is relatively quick and leads to long-run equilibrium.
H. Post Diagnostics Check
Following the evaluation of the VECM model, the next step is to determine whether the appropriate model chosen
provides efficient data estimates and is well-specified.
Various post-estimation tests are performed to check for
misspecification and stability issues, as well as the problem of
serial correlation, among other things. All diagnostic checks
show that the model is stable, with no signs and symptoms of
autocorrelation, normality, or heteroskedasticity.
As a result, the model is well-designed and free of
any econometric issues and mistakes. The inverse roots of the
AR polynomial stability condition test display that no points
are outside the circle, indicating that the VECM model is
stable and the results are correct and valid. Based on the serial
correlation test results, the probability values show no
evidence of serial correlation among residuals. The null
hypothesis of no serial correlation is accepted at a 5 percent level of significance.The heteroskedasticity test
revealed that there is no heteroskedasticity in the model.
a) Stability Test
The VECM was tested for stability using an eigenvalue stability test, which revealed that the system is stable. K – r
unit eigenvalues are contained in the companion matrix of the
VECM with K endogenous variables and r cointegrating
equations. The moduli of the remaining r eigenvalues are
strictly less than one if the process is stable. It is difficult to
determine whether the eigenvalues' moduli are too close 1 and
similar because there is no general distribution theory for
moduli. The general rule in this test is that there would be a
sustainability problem if some of the remaining calculated
modules were very close to one. Eigenvalues appear in table
19 below. Only one of the eigenvalues is one, while the other
computed modules are not close to one.
The results show that there are about 0.338 of real roots.
There is no distribution theory that can be used to determine
how close this root is. Nonetheless, the root 0.339 supported earlier estimate concluded that the predicted
cointegration equation was most likely not stationary. The
eigenvalue graph reveals that the remaining eigenvalues are
not found near the unit circle. The stability check does not
indicate that the model was incorrectly specified. The graph
that accompanies figure 4 shows that all eigenvalues are
within the unit circle. The companion matrix has (5-1) unit
eigenvalues for a five-variable model with one cointegration
relationship. For stability, the remaining eigenvalue
coefficients must be exactly less than unity. Below, diagnostic
checks will be carried out on heteroskedasticity, the
autocorrelation of residuals, and normality.
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 391
Table 19: Stability Test
b) LM Test of Autocorrelation
The Lagrange-Multiplier test is then used to check and
look for autocorrelation in the residuals of the VECM.
The null hypothesis of no autocorrection occurs at the optimal
lag selection. After evaluating the VECM model, we test
whether the residuals of this model are white noise or not. If
autocorrelation is observed between the residues, then it is implied that some information is left out of the model, such as
insufficient lags. The Lagrange-Multiplier (LM) test detects
serial correlation of residues up to a specified order of lag,
which should be the same as the corresponding VAR. The
result of the LM test at the 5 percent level is shown in Table
17. The null hypothesis, that there is no autocorrelation in the
residuals for any of the orders investigated, cannot be rejected.
As a result, no evidence of model misspecification can be
found in this test.
c) Multivariate Heteroskedasticity Test
Statistical evidence shows that the system is
homoskedastic or free of heteroskedasticity by checking the
Fig. 5: Stability Test
test in table 20 for the multivariate ARCH effect because the p-value under the null hypothesis is nonsignificant up to
the order of lagged 2.
d) Test for Normality of Residuals
Finally, an analysis of model disturbances is required. The test aims to check whether the residual has a normality
distribution or not. Jarque-Bera test, Kurtosis test,
andSkewness testcomputes series test statistics for H0, which
aid in confirming the VECM disturbances are normally
distributed. Failure to reject H0 indicates that the model is
incorrectly specified or misspecified. However, the lack of
residual normality does not render or cause the cointegration
tests and VECM invalid. The results of table 21, 22, and 23
shows that the model is not misspecified because null
hypothesis H0 is accepted. Thus the results regarding the
variables which have an effect on the container throughput could be accepted. Kurtosis test statistics do not reject H0 that
the terms of disturbance have kurtosis and skewness test
consistent with normality.
Table 19: Lagrange-Multiplier Test Table 20: Heteroskedasticity Test
.
The VECM specification imposes 4 unit moduli.
-.3340626 .334063
-.4764774 - .05195964i .479302
-.4764774 + .05195964i .479302
.6027718 .602772
.3384209 - .6716343i .752078
.3384209 + .6716343i .752078
1 1
1 1
1 1
1 1
Eigenvalue Modulus
Eigenvalue stability condition
. vecstable, graph
H0: no autocorrelation at lag order
2 21.6318 25 0.65691
1 21.5027 25 0.66427
lag chi2 df Prob > chi2
Lagrange-multiplier test
H0: no autocorrelation at lag order
2 21.6318 25 0.65691
1 21.5027 25 0.66427
lag chi2 df Prob > chi2
Lagrange-multiplier test
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 392
Table 21: Jarque-Bera Test Table 22: Kurtosis Test
Table 23: Skewness Test
Source: Author’s calculation
V. DISCUSSION
This study has discovered many interesting findings of
the factors affecting the maritime transport markets by using
applying the VECM model of time series econometric analysis to investigate the container
throughput of Cambodia's port and the problem that needs to
be taken action and solved for better improvement to
Cambodia's international ports to support the higher increases
of the container throughput traffics and demands in Cambodia
in long run plan and project of maritime transport as well as
strategy development of port and shipping market.The
findings of analysis and estimation in the also reveal many
interesting findings, which are discussed below;
A. Number of Ship Call and Container Throughput
The number of ship called has a significant and positive
impact on the combined container throughput of PAS and
PPAP. This result demonstrates that the flow of traffic
demand in ports explains Cambodia's port throughput and is
related to port performance and efficiency from an increase in number of ships that called at ports. It means that the port
authorities of PAS and PPAP are paying attention to and
focusing on the number of ship called at ports in order to meet
the increased demand for container throughput in ports,
because port performance is measured in terms of the number
of containers throughput. This is consistent with JICA's
master plan, which studied master plan ports in both PAS and
PPAS to have deep-sea port construction expand the large port
terminals and port facilities to support the higher container
throughput from the increased frequency of ship call in the
long run. The total number of ship call exhibits a positive sign
of association for statistically significant variables. It is positive for port throughput; ship call represent a lot of
opportunity and demand for the global shipping network, as
shown in Japan, indicating a congested condition at the port
level. The call frequency will help to support this
circumstance. The higher shipping frequency improves the
more efficiency of terminal operations. The researchers also
found that significant container ship call in Cambodia
improved container traffic throughput due to a significant
increase in the average volume carried by each ship. The
volume of containers passing through a terminal or port is
influenced by the number of ship call. If the number of ship
call is increased, it will be more appealing to exporters and
importers. Furthermore, suppose the port or terminal is
capable of providing higher quality and additional services to
commercial shipping. In that case, shippers will determine container cargo flows, thereby providing a competitive port.
Besides, liner shipping agreements and ship or vessel upsizing
strengthen the relationship between container shipping lines
and container terminals. As a result, shipping alliances can
choose to come to a port that will benefit them the most in
terms of deployed capacity, ports of call, network structure,
and port choice, and so on.
B. Import and Container Throughput
The findings result of import has a significant and negative
relationship with the container throughput of ports PAS and
PPAP. This result proves that Cambodia's port throughput in
the long term of the plan is explained by a decrease in imports
and is related to improved higher port efficiency, port
performance and terminal traffic efficiency. This negative
impact can be described as the fact that most imports or seaborne trade of imported goods from other countries' trading
partners to Cambodia will be reduced due to the high cost of
maritime shipping transportation and logistics and other
surcharges. Imported goods are becoming more expensive,
resulting in a decrease in import demand.The customs charge,
tariff, and port due are still the most significant impediments
to improving future import balance. As a result, other modes
of transport, such as land and air, which are less expensive
ALL 4.424 10 0.92618
D_lngdppc 0.610 2 0.73704
D_lnimp 0.450 2 0.79843
D_lnexp 0.906 2 0.63568
D_lnnsc 0.322 2 0.85139
D_lncont 2.136 2 0.34367
Equation chi2 df Prob > chi2
Jarque-Bera test
ALL 2.144 5 0.82891
D_lngdppc 2.5828 0.174 1 0.67654
D_lnimp 2.8556 0.021 1 0.88516
D_lnexp 2.6695 0.109 1 0.74101
D_lnnsc 2.5146 0.236 1 0.62739
D_lncont 4.2665 1.604 1 0.20534
Equation Kurtosis chi2 df Prob > chi2
Kurtosis test
ALL 2.281 5 0.80910
D_lngdppc .33022 0.436 1 0.50897
D_lnimp -.32763 0.429 1 0.51231
D_lnexp -.44634 0.797 1 0.37203
D_lnnsc -.14675 0.086 1 0.76914
D_lncont -.36473 0.532 1 0.46572
Equation Skewness chi2 df Prob > chi2
Skewness test
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 393
than shipping by sea, will be phased out. The variable import
coefficient has a negative sign, indicating that Cambodia appeared to be diverting trade. It demonstrates a decline in
imports via shipping from ASEAN member countries or
trading partners. However, despite having a negative sign, the
estimated coefficient on import is significant, implying that
there is insufficient evidence that maritime transport shipping
of imported goods will affect Cambodia's imports in the
future. The existence of a direct relationship between import
and container throughput of international ports of Cambodia,
in other words, implied that a 1 percent increase in import
would lead to a negatively long-run effect on the performance
by 1.5 percent. This result contradicted the expected increase
in positively affected imports.
Several economic facilitation programs will be
implemented to address the economic problems of
international trade, the reduction and/or exemption of custom tariffs for imported goods, the compensation of tax and duty
paid on domestic goods for export, the reduction of import
duty for parts and spare parts for all types of automobiles,
including the reduction and/or exemption of income tax
related to international trade, and the reduction of the cost of
export production. This crucial policy framework and
mechanism will be promote and help define as the tools of
economic sector in order to measure the direct and indirect
impact of international trade promotion, and trade
exchanges.It should be noted that the above-mentioned
economic facilitation programs aim to lower a component of
import and export costs. The government, on the other hand, will be forced to increase national spending in the trade sector.
C. Export and Container Throughput
At the 5 percent level, the estimated coefficient export is
positive and statistically significant, indicating that export influences the demand for container throughput at the
combined Cambodian ports of PAS and PPAP. Export trade
volumes will rise in parallel with economic growth. Cambodia
is expected to move 4.1 times more products in 2030 than it
did in 2018. Maritime transport sectors and ports will increase
cargo and port throughput due to the increase in domestic
commodity exports. This significant and substantial
growth will require the development of both soft and hard
logistics infrastructure. Maritime transport policy is becoming
more efficient and effective to handle and process the
increased export volumes and increased export volumes and a wider range of import and export values. Cambodia must
improve its competitiveness to ensure that its exports-to-GDP
ratio remains at or near current levels in order to sustain
strongly significant economic growth in the medium and long
term. Cambodia should look into efficient and effective ways
to reduce logistics and supply chain bottlenecks in order to
improve service reliability and keep its international market
competitiveness. Because exports are satisfied with tariff-free
preferences in the markets of the US and EU, and are not
heavily influenced by ocean freight fees, rates, and
surcharges, reducing logistics costs, charges, and delays will
be the main strategic ways for Cambodia to further strengthen and maintain its high economic growth rates over the next 15
years.
Similarly, an increase in demand for imported goods by a
trading partner may result in higher demand created by an increase in the countries' exports. Some of the higher partners'
demand may increase Cambodia's exports, resulting in an
increase in maritime transport activities through Cambodia's
busiest ports, resulting in an increase in container throughput.
Export cargo is transported via the international trade ports of
PAS and PPAP. Some commodities are shipped by road from
or to Thailand, Laos, and Vietnam, as well as by air. Export
volume through both international ports has been increasing,
with the growth rate of containers through Phnom Penh Port
being particularly remarkable. The International distribution
ports will expand in the future, and export container and bulk
cargo will increase as a result of rational role sharing between land and sea transportation. PAS and PPAP have served as
international distribution centers in Cambodia and will
continue to do so in the future. Foreign trade through private
ports, on the other hand, is increasing, and private port
capacity is expanding rapidly. In the future, the private sector
intends to construct additional large ports. The roles of these
ports must be rationally assigned, taking into account the
unique characteristics of each port as well as the efficient use
of resources. Cambodian ports will improve their overall
functions through cooperation and competition among
themselves.
D. GDP Per capita and Container Throughput
GDP per capita has a negative sign and has a significant
impact on the demand for port throughput, which is
considered the port output and efficiency. In terms of the economic variable, container throughput and economic
growth are fundamentally linked with a tightening relationship
that is similar to China and generates a lot of container
throughput or port cargo like several other large developing
countries. Nonetheless, GDP per capita is lower than average,
resulting in a negative estimate. According to Vanoutrive,
Thomas (2010), a study linking economic activities and port
throughput in Belgium discovered very similar results that
demonstrated the correlation. The reason for this has already
been discussed, and it is due to Cambodia as a small country.
The relationship between port throughput and neighboring countries' GDP is frequently stronger than or equal to the
relationship between port throughput and the GDP of the
country in which the port is located.
More trade, particularly in intermediate products, is accompanied by regional or global supply chains, according to
Goldfarb and Beckman (2007); Robinson (2002); Goldfarb
and Thériault 2008). Because values are created at various
points along the supply chains, the relationship between what
is transported via container ship and port and its economic
impact is more complex. Musso et al. (2002) provides another
plausible explanation of outcome and result described how the
national economy was becoming increasingly irrelevant to
port success. This means that external economies, foreign
countries are becoming more important than national or local
economies. The effects of GDP per capita on port
performance or container throughput yields may be minor in the long run. The economic crisis, illness, or shock will all
have a greater impact, and GDP per capita has even become
negatively and significantly related on the demand for
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 394
throughput of ports. According to Wildenboer, E. (2015)
shown that when per capita GDP rises by 1000, container falls by 451,608 TEU. This parameter has a large negative impact
on container throughput. This is a remarkable discovery
because a significant positive relationship between per capita
income and national GDP on the one hand, and port
performance on the other, can be predicted.
VI. CONCLUSION
Based on the estimated results and discussions, it is
possible to conclude that the endogenous variable was container throughput for Cambodia's international ports,
namely the PAS and PPAP ports. Simultaneously, the
exogenous variables are the number of shipcall, export,
import, and GDP per capita, which are all in natural
logarithms. To estimate research work, the cointegration
equation and the VECM are used. Meanwhile, it is discovered
that the series used in this analysis is nonstationary at the level
but stationary at the first differencing.A sequential likelihood
ratio test was performed on the rank of the estimated
parameter matrix from the VECM model to identify the
number of cointegration relations. Recently, the IRF was used to determine whether or not there is any shock in the response
of one factor to another.The empirical analysis of this study
reveals the best fit of the model by developing the long-run
equilibrium relationship advancement of the actual data time
series data, as evidenced by appropriate estimation findings of
results. The study also further examines the long-run causal
relationships among the variables influencing container
throughput in time series data. The statistical and econometric
evidence revealed that the estimating series has a statistically
and significantly long-run equilibrium relationship, which
affects Cambodia's overall container throughput.
Besides that, it is critical to emphasize that the VECM
results are not based on the short-run granger-causality
relationship that runs from GDP per capita, ship call, import
and export, to container throughput. According to the estimation results of the IRF test, the shock response of
independent variables to container throughput disappears after
a while. Furthermore, the container throughput at Cambodia's
ports generated by VECM yielded satisfactory results.The
model was appropriate, and other methods of measuring tests
were considered in order to evaluate the model's goodness and
accuracy. The PP and ADF results for unit root tests
demonstrate that all series are nonstationary at the level but
stationary in the first differencing, i.e. they are integrated with
the order one I (1). The research then keeps moving on to
examine the long-run equilibrium relationship among
exogenous variables using Johansen's multivariate cointegration test. The findings revealed the presence of a
cointegration vector between these variables. Various
diagnostic and post-estimation tests are performed to
investigate misspecification issues as well as the model's
stability and robustness in order to obtain the best possible
results and correctly confirm them. The stability check shows
that our model is not misspecified or incorrectly
specified. Based on the Lagrange-Multiplier for LM test
results, we cannot reject the null hypothesis, which states that
there is no serial correlation or autocorrelation the residuals
for any of the orders studied. As a result, the results of the
tests show that there is no evidence of model misspecification.
Our model does not have misspecification, according to Jarque-Bera, Kurtosis, and Skewness test results. For all
equations, the null hypothesis for the normality of distributed
convergence to recover the long-run equilibrium position is
not rejected. As a result, the model is satisfied and appropriate
in the research and analysis. As a matter of conclusion, it is
possible to conclude that the number of ship call has a
statistically and significantly positive effect on container
throughput. The coefficient of a number of ship call is
1.82213, indicating that it is more elastic (greater than 1). It
shows that a 100 percent rise in the number of ship called is
likely to increase container throughput by 182.22 percent in
the long run in a positive relationship to Cambodia's port throughput. An rise in the number of ships calling at
Cambodian ports would boost maritime sector activities such
as loaded or unloaded cargo transportation and the frequency
of shipping traffic at the ports.As a result, container
throughput would increase. Export has a statistically
significant impact on container throughput and has a positive
effect on Cambodian container throughput.The coefficient
value is explained by 1.00068, which is greater than 1 and is
considered elastic. When exports increase by 100 percent,
container throughput increases by more than 100.068 percent.
As a result, Cambodia should prioritize export promotion policies in order to increase seaborne trade and export cargoes
of final products from its countries, allowing maritime time
transport activities to become more interesting and busy as the
tools and measure in TEUs of container put throughput at
port.
The findings of import also indicated and thus confirmed
a long-run negative relationship between import and container
throughput. The coefficient value is produced by minus
1.51536, which is greater than 1 of absolute value and
considered elastic, implying that a 100 percent increase in
import will reduce the impact of nearly 151.53 percent of
container throughput long-run negative relationship. The
findings revealed a long-run negative relationship between
import and demand for container throughput because imports
of goods are valued at their free on board market prices at foreign ports, which do not include Cambodian import duties,
ocean freight, maritime transport services such as insurance,
travel expense and maritime transport freight paid to foreign
carriers or firm that need to be shipped because imported
goods are not produced in Cambodia, so the cost will be more
expensive.
Also, GDP per capita and container throughput have a
significant effect on and a negative long-run relationship with
each other in line with the studies conducted by OECD (2014)
on time efficiency at world port container,. The value of the
coefficient is explained by minus 1.00532, which is greater
than 1 and is considered elastic in absolute value. The result
indicated a 100 percent increase in GDP per capita; however,
with a negative long run relationship, container throughput is
reduced by more than 100.53 percent. This can lead to the
conclusion that international economies and the hinterland, neighboring countries, or the rest of the world are given more
weight than national or local economies. This demonstrates
that the national economy is becoming less and less essential
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 395
for port performance as a result of the economic crisis,
random shocks, and diseases in the country, such as SARs in Hong Kong and Covid-19 of the global pandemic, all of
which reduce a country's GDP and GDP per capita. Despite
this, the higher container port cargo throughput was still
significantly higher.With the best results, it can now conclude
that the coefficient values are all elastic. As a result, it is
concluded that container throughput is relatively more elastic
in terms of the number of ship call than in terms of import,
export, and GDP per capita, and there are also long-run
equilibrium relationships. In the short run, the granger
causality test revealed no causal relationship among container
throughput and exogenous variables like GDP per capita
import and export, and ship call.
In the short run, the granger causality test revealed no
causal relationship between container throughput and
exogenous variables like GDP per capita import and export, and ship call. This finding was consistently responsible for the
long analysis, which is explained by the fact that
the joint variables of GDP per capita, number of ship call,
import, and export do not granger-cause container throughput
is rejected.
VII. POLICY IMPLICATION AND RECOMMENDATION
Cambodia's government ports, MPWT, and port
authority should develop, design, and implement the best port policy and strategy planning, such as developing future
terminals, infrastructure, container ports, and facilities, in
order to provide better maritime transport service, quality port
services, and shippers and liner shipping
companies.Moreover, due to the capacity of the existing
container port, the building of new ports to meet the future
demand will increasingly be unable to handle such massive
container demand (Syafi'i et al., 2005; JICA, 2007).The
government and MPWT should consider developing and
formulating a maritime transport strategic policy to promote
long-term economic growth and sustainability. The findings of this study suggested that port authorities may have to make
all necessary decisions to invest in developing infrastructures
of ports and port facilities, transportation, container terminals,
and logistic connectivity, among other things, because
increased productivity of port facilities will lead to increased
demand for container throughput or volume of throughput in
the future.Port capacity must be increased in order to meet and
forecast rising demand. As a result, the robust analysis and
estimated results of the findings found in this research greatly
assist the government, MPWT, and port authorities in making
better decisions that could provide appropriate values for
meeting the increasing demand of throughput of Cambodian ports and investment, finance, and operation in the future to
handle and sustain the higher demand of throughput.
VIII. DECLARATION
The author declares that there are no conflicts of interest
in the publication of this paper. This research is not supported
by any external funding or grants from public, non-profit, or
commercial organizations. Errors and omissions, if any, are solely the author's responsibility.
IX. ACKNOWLEDGEMENT
The author wishes to thank the anonymous peer-
reviewers and editors for their insightful comments, which
helped improve this scientific journal from its original form,
as well as Dr. Kang Sovannara, Dr. NhemSareth, and Dr. Kim
ChomroeunVuthy. My heartfelt gratitude goes to Southern Denmark University and the MERE group scholars, together
with Dr. NielsVestergaard, Dr. Eva Roth, Dr. Brooks Kaiser,
Dr. Melina Kourantidou, and Dr. Lars RavnJonsen.
REFERENCES
[1.] Andrea Mainardi, (2016). Forecasting cargo throughput
in Portuguese ports.Dissertaçãoparaobtenção do Grau de
Mestreem.
[2.] ArjunMakhecha, (2015-2016). Analysis of the determinants of container throughput of the major ports
in the Hamburg Le Havre range.Erasmus University
Rotterdam.MSc in Maritime Economics and Logistics.
[3.] Armstrong, J. and Brodie, R., (1999). Forecasting for
marketing, in Hooley, G. and Hussey, M. (Eds),
Quantitative Methods in Marketing, 2nd edition, London
International Thompson Business Press, London, pp. 92-
119.
[4.] Banerjee, A., Dolado, J.J., Galbraith, J.W., and Hendry,
D.F., (1993). Cointegration, Error Correction, and the
Econometric Analysis of Non-stationary Data, Oxford University Press, Oxford.
[5.] Chou, C. C., Chu, C.W. and Liang, G.S., (2008). A
modified regression model for forecasting the volume of
Taiwan’s import containers.Mathematical and Computer
Modelling, Vol. 47 Nos 9/10, 797-807.
[6.] Chou, C. C., Chu, C.-W., & Liang, G.-S. (2008a). A
modified regression model for forecasting the volumes
of Taiwan’s import containers. Mathematical and
Computer Modelling, 47(9–10), 797–807.
[7.] Chou, C. C., Chu, C.-W., & Liang, G.-S. (2008b). A
modified regression model for forecasting the volumes of Taiwan’s import containers. Mathematical and
Computer Modelling, 47(9–10), 797–807.
[8.] Chung, K. C., (1993). Port performance indicators.World
Bank Infrastructure Notes.
[9.] Cruz, M. R. P., Ferreira, J. J., &Azevedo, S. G., (2013).
Key factors of seaport competitiveness based on the
stakeholder perspective: An Analytic Hierarchy Process
(AHP) model. Maritime Economics & Logistics, 15(4),
416-443.
[10.] Cullinane, K., (2004). The Container Shipping Industry
and The Impact of China's Accession to the WTO.
Research in Transportation Economics, Vol. 12, No.221-245.
[11.] De Monie, G., (1987). Measuring and Evaluating Port
Performance and Productivity, UNCTAD Monographs
on Port Management, Geneva.
[12.] Deng, P., Lu, S., & Xiao, H., (2013). Evaluation of the
relevance measure between ports and regional economy.
Transport Policy, 27, 123-133.
[13.] Dickey, D.A., and Fuller, W.A.., (1979). Distribution of
Estimators for Autoregressive Time Series With A Unit
Root, Journal of the American and Statistics Society, 74,
pp. 427-431.
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 396
[14.] Drewry Shipping Consultants, (2010).Container Market
(2010/11).Annual Review and Forecast Incorporating the Container Forecaster.
[15.] Enders, W., (2004). Applied Econometric Time Series,
Second Edition. John Wiley & Sons Inc, New York.
[16.] Enders, W., (2010). Applied Econometric Time
Series.Third edition, John Wiley & Sons Inc, New York.
[17.] Enders, W., (2015). Applied Econometric Time Series.
New York: John Wiley, 1–63.
[18.] Engle, R. F. and Granger, C. W., (1991). Introduction. In
Engle, R. F. and Granger, Long-run economic
relationships: Readings in co-integration, pages 1–16.
[19.] Engle, R.F. & Granger, C.W.J., (1987). Co-integration
and error correction: representation estimation, and testing. Economectrica, 55, 251-276.
[20.] Engle, R.F., and Yoo, B.S., (1991). Cointegrated
Economic Time Series: An Overview with New
Results.Oxford University Press, Oxford, U.K.
[21.] Goldfarb, D. and Beckman, K., (2007). Canada's
Changing Role in Global Supply Chains, The
Conference Board of Canada, Ottawa.
[22.] Goldfarb, D. and Thériault, L., (2008). Canada's
"Missing" Trade With Asia. The Conference Board of
Canada, Ottawa.
[23.] Gonzalo, J., (1994). Five Alternative Methods of Estimating Long-run Equilibrium Relationships, Journal
of Econometrics, 60, pp. 203-233.
[24.] Gosasang V., Chandraprakaikul W., Kiattisin., (2011). A
Comparison of Traditional and Neural Networks
Forecasting Techniques for Container Throughput at
Bangkok Port. The Asian Journal of Shipping and
Logistics. Vol.27 No.3, 463-482.
[25.] Gujarati, (2008). Basic Economterics, MCGraw-Hill,
New York, USA.
[26.] Gujarati, (2012). Basic Economterics, MCGraw-Hill,
New York, USA.
[27.] Hakim, M. M., &Merkert, R., (2016). The causal relationship between air transport and economic growth:
Empirical evidence from South Asia. Journal of
Transport Geography, 56, 120–127.
[28.] Imai, A., Shintani, K., & Papadimitriou, S., (2009).
Multi-port vs. Hub-and-Spoke port calls by
containerships. Transportation Research Part E:
Logistics and Transportation Review, 45(5), 740-757.
[29.] Jiang, J., Wang, H. Y., & Yang, Z., (2007). Econometric
analysis based on the throughput of container and its
main influential factors. Journal of Dalian Maritime
University, 2, 83-86. [30.] Jiaweia, Z., Lizhib, X., Xuna, Z., &Shouyanga, W.,
(2012). Forecasting Container Throughput within Multi-
Port Region Using CDE-MPR Methodology: the Case of
PRD Region.
[31.] JICA, (2007). The Study on the Master Plan for
Maritime and Port and Sectors in the Kingdom of
Cambodia.
[32.] JICA, (2017). Preparatory Survey for Phnom Penh
Autonomous Port: New Container Terminal Special
Economic Zone and Associated Facilities Construction
Project in Kingdom of Cambodia. Final Report.
[33.] JICA, (2017). Preparatory Survey for Sihanouk Ville
Port: New Container Terminal Development Project. Final Report.
[34.] Johansen,S.,(1988). Tatistical Analysis of Cointegrating
Vectors, Journal of Economic Dynamic and control, 12,
pp. 231-254.
[35.] Johansen, S., (1991). Estimation and Hypothesis Testing
of Cointegrated Vectors in Gaussian Vector
Autoregressive Models, Econometrica, 59, pp. 1551-
1580.
[36.] Johansen, S., Julius, K., (1994). Identification of The
Long-run and The Short-run Structure: An Application
to The ISLM Model, Journal of Econometrics, 63, pp. 7-
36. [37.] Johansen, S., Juselius, K., (1990). Maximum Likelihood
Estimation and Inference on Cointegration With
Applications to the Demand for Money. Oxford Bulletin
of Economics and Statistics, 52(2), 169-210.
[38.] Keck, A., A. Raubold, and A. Truppia. (2009).
Forecasting International Trade: A Time Series
Approach. OECD Journal: Journal of Business Cycle
Measurement and Analysis, 157–176.
[39.] Kutin, N., Nguyen, T. T., &Vallée, T. (2017). Relative
efficiencies of ASEAN container ports based on data
envelopment analysis. Asian Journal of Shipping and Logistics, 33(2), 67-77.
[40.] Langen, P. W., (2003). Forecasting container
throughput: a method and implications for port planning.
Journal of International Logistics and Trade, 1(1), 29-39.
[41.] Langen, P. W., Van Meijeren, J., &Tavasszy, L. A.,
(2012). Combining Models and Commodity Chain
Research for Making Long-Term Projections of Port
Throughput: an Application to the HamburgLe Havre
Range. European Journal of Transport and Infrastructure
Research, 12(3).
[42.] Langen, P. W., Van Meijeren, J., &Tavasszy, L. A.,
(2012). Combining Models and Commodity Chain Research for Making Long-Term Projections of Port
Throughput: an Application to the HamburgLe Havre
Range. European Journal of Transport and Infrastructure
Research, 12(3).
[43.] Langen, P., Nijdam, M., & Horst, M., (2007). New
indicators to measure port performance. Journal of
Maritime Research, IV(1), 23-36.
[44.] Langen, P., Nijdam, M., & Horst, M., (2007). New
indicators to measure port performance. Journal of
Maritime Research, IV(1), 23-36.
[45.] Levinson, M., (2006). How the shipping container made the world smaller and the world economy bigger,
Princeton University Press.
[46.] Liu Bin, Zhao Yichuan, Chen Yong., (2002). The
correlation analysis on port container throughput and
main macroeconomic indices, World Shipping, 2002;
Vol.25, No.5, pp.50-51.
[47.] Liu, L., & Park, G.K., (2011). Empirical Analysis of
Influence Factors to Container Throughput in Korea and
China Ports. The Asian Journal of Shipping and
Logistics, 27(2), 279–303.
[48.] Lutkepohl, H., (1991). Introduction to multiple time series analysis, Springer-Verlag.
Volume 6, Issue 11, November – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IJISRT21NOV450 www.ijisrt.com 397
[49.] Meersman, H. and Van De Voorde, E., (2008). The
Relationship between Economic Activity and Freight Transport, pp. 69-92. In M. Ben-Akiva, H. Meersman,
and E. Van De Voorde (eds.), Recent Developments in
Transport Modelling-Lessons for the Freight Sector.
Emerald, Bingley, UK.
[50.] Nakano Hiromichi., (2004). The new observation on
container transportation demand forecasting, Quarterly
69, Vol.1,2, 8.
[51.] Nam, H.S., and Song, D.W., (2011). Defining maritime
logistics hub and its implications for container port,
Maritime policy & Management, Vol. 38, pp. 269-292.
[52.] Ng, A. K. Y., and J. J. Liu., (2014). Port-Focal Logistics
and Global Supply Chains. Basingstoke: Palgrave Macmillan.
[53.] OECD, (2014). César Ducruet, HidekazuItoh, Olaf Merk
- Discussion Paper 2014-08 - OECD/ITF.
[54.] Paflioti, P., Vitsounis, T. K., Teye, C., Bell, M. G.,
&Tsamourgelis, I. (2017). Box dynamics: A sectoral
approach to analyse containerized port throughput
interdependencies. Transportation Research Part A:
Policy and Practice, 106, 396-413.
[55.] Pallis, A. A., T. K. Vitsounis, P. W. De Langen, and T.
E. Notteboom. 2011. Port Economics, Policy and
Management - Content Classification and Survey. Transport Reviews 31 (4): 445–471.
[56.] Peng, W.Y. and Chu, C.W., (2009). A comparison of
univariate methods for forecasting container throughput
volumes. Mathematical and Computer Modelling, Vol.
50 Nos 7/8, 1045-1057.
[57.] Perron, P., (1989). The great crash, the oil price shock,
and the unit root hypothesis. Econometrica: Journal of
the Econometric Society, 1361–1401.
[58.] Poitras, G., Tongzon, J., & Li, H., (1996). Measuring
port efficiency: An application of Data Envelopment
Analysis.Working Paper, National University of
Singapore. [59.] PPAP, (2015). Annual Report in Khmer Version.
[60.] PPAP, (2015). Disclosure Document for Initial Public
offering of Equity Securities. Phnom Penh Autonomous
Port. License number: 0127 MOC/D/REG 05. Issued by
SECC.
[61.] PAS, (2016). Annual Report in Khmer Version.
[62.] PPAP, (2017) Annual Report in Khmer Version.
[63.] Qiuhong, Z. (2009). Application of grey model in
forecasting the port of Qinhuangdao’s throughput.2009
IITA International Conference on Services Science,
Management and Engineering, 57–60. [64.] Rashed, Y., Meersman, H., Van de Voorde, E.,
&Vanelslander, T. (2013). A univariate analysis: Short-
term forecasts of container throughput in the port of
Antwerp.
[65.] Robinson, R., (2002). Ports are elements in value-driven
chain systems: the new paradigm. Maritime Policy
Management 29(3): 241–255.
[66.] Seabrooke, W., Hui, E. C., Lam, W. H., & Wong, G. K.,
(2003). Forecasting cargo growth and regional role of
the port of Hong Kong. Cities, 20(1), 51-64.
[67.] Slack, B., (1985). Containerisation and inter-port competition. Maritime Policy & Management, Vol. 12,
No.4, 293-304.
[68.] Song, D. W., & Han, C. H., (2004). An econometric
approach to performance determinants of Asian container terminals. International Journal of Transport
Economics/Rivistainternazionale di
economiadeitrasporti, 39-53.
[69.] Stock J.H. and Watson, M.W., (2012). Generalized
Shrinkage Methods for Forecasting Using Many
Predictors. Journal of Business & Economic Statistics,
30(4), 481-493.
[70.] Syafi’I, Kuroda, K., and Takebayashi, M., (2005).
Forecasting the demand of container throughput in
Indonesia. Memoirs of Construction Engineering
Research Institute, 47(47):69–78.
[71.] Syafi’I. (2009). Multivariate autoregressive model for forecasting the demand of container throughput in
Indonesia. Media TeknikSipil, 6(2):129–136.
[72.] Syafi’I., (2006). Multivariate autoregressive model for
forecasting the demand of container throughput in
Indonesia. Media Teknik Sipil, 129-134.
[73.] Tongzon, J. (2001) Efficiency Measurement of Selected
Australian and Other International Ports Using Data
Envelopment Analysis. Transportation Research Part A:
Policy and Practice, 35, 107-122.
[74.] Tongzon, J.L., (1995). Determinants of port performance
and efficiency. Transportation Research Part A: Policy and Practice, 29(3), 245-252.
[75.] Unctad, (1976). Port performance indicators. New York:
United Nations.
[76.] Unctad, (2004). Review of Maritime Transport
2004.New York & Geneva, USA.
[77.] Unctad, (2013). Review of Maritime Transport
2013.New York & Geneva, USA.
[78.] Unctad, (2018). Review of Maritime Transport 2017.
New York & Geneva: USA.
[79.] Vanoutrive, T. (2010), “Exploring the Link between Port
Throughput and Economic Activity: Some Comments on
Space-and-Time Related Issues”, 50th Anniversary European Congress of the Regional Science Association
International (ERSA) Sustainable Regional Growth and
Development in the Creative Knowledge Economy 19th
– 23rd August 2010, Jönköping, Sweden.
[80.] Wildenboer, E., (2015). The relation between port
performance and economic development.
[81.] Woo, S.H., Pettit, S. & Beresford, A. K., (2011a). Port
Evolution and Performance in Changing Logistics
Environments. Maritime Economics & Logistics, 13,
250-277.
[82.] Wooldridge, J., (2010) Introductory Econometrics: A Modern Approach.
[83.] Yap, W.Y.,& Lam, J.S.L., (2008). Competition for
transhipment containers by major ports in Southeast
Asia: slot capacity analysis. Maritime Policy &
Management, Vol. 35, No.1, 89-101.
[84.] Zhu Xiaomeng, (2014). A study on correlation model of
seaport container throughput and gross domestic
product, Dissertation, Dalian Maritime University.